You are on page 1of 14

Journal of Environmental Management 247 (2019) 371–384

Contents lists available at ScienceDirect

Journal of Environmental Management


journal homepage: www.elsevier.com/locate/jenvman

Research article

Methodology for improving reliability of river discharge measurements T


E.S. Bekri , P.C. Yannopoulos, P. Economou

Environmental Engineering Laboratory, Department of Civil Engineering, University of Patras, 265 04, Patras, Greece

ARTICLE INFO ABSTRACT

Keywords: Programs to monitor water characteristics are undertaken to identify any possible water pollution of a river. To
River flowrate measurement compute reliable water pollution loads, accurate river discharge and pollutant/tracer concentrations are re-
Water pollution load quired. When pollutants/tracers are measured with sufficient precision, the accuracy of river discharge mea-
Measurement error-correction surements becomes the most critical parameter of the pollutant load computation, as well as the largest error
Nonlinear optimization
source. The absence of permanent measuring equipment in many rivers is a common difficulty for the im-
Water volume conservation
plementation of monitoring programs. Alternatively, quick measurement methods which are low in cost and
Tracer mass conservation
reliability (e.g. floats) are often employed to get an estimate of river discharges, when there are budgetary and
time restrictions. In this paper, an original technique, mainly for use in ungauged rivers, is proposed for cor-
recting river discharge measurements which have low levels of accuracy; this in turn would correct pollutant
concentrations. A nonlinear optimization problem is developed based on water volume and pollutant mass
conservation principles for river balance nodes, taking into consideration non-measurable latent quantities.
Parallel measurements of discharge and tracers for representative cross-sections of a river and its tributaries are
required. The measurement conditions should refer to the steady-state hydraulic conditions usually prevailing in
the flow under consideration. In order to test the reliability of the method, a virtual river example is built,
defining the real values of water characteristics and generating measurement sets via Monte-Carlo simulations
combining random and systematic errors. For more than 92% of the generated measurement sets, the proposed
technique results in a successful and acceptable correction for the total of the measured cross-sections. Finally,
the method is applied to a real river and the measurements are corrected successfully.

1. Introduction error in measurements and, on the other side, to develop and integrate
techniques and procedures to correct the measured values, or even try
One of the key stages of river basin management is the monitoring to derive more reliable values.
of water quantity and quality in order to: establish a coherent and A plethora of river discharge measurement methods has been de-
comprehensive overview of water status; identify changes or trends in veloped and described (WMO, 1980; ISO, 1981, 1988; Rantz, 1982;
water quality and quantity; and to assess remediation or preventive Kinori and Mevorach, 1984; White, 1988; Müller, 1988; Fujita, 2017).
measures within each river basin district. According to the European The choice of each method is determined by the features of the river
Water Framework Directive 2000/60/EC (WFD, 2000), river mon- and the available measuring apparatus. For a holistic and complete
itoring programs should determine the mass load as well as the level of picture of the whole river status, including its tributaries, quantitative
predefined pollutants. The mass load of a pollutant/natural tracer at a and qualitative characteristics should be measured in tandem at sui-
selected river cross-section is indirectly estimated by the combination tably-chosen cross-sections, which embody the whole river. To achieve
of parallel measurements of water discharge and pollutant concentra- this, fixed discharge measurement arrangements should be present, as
tion; since its calculation results from the multiplication of these two well as continuously-functioning automatic samplers for the measure-
measurements. It is worth noting that numerous measurements and ment of pollutant concentrations. River discharge is ideally estimated
samples are needed to accurately and reliably capture the true value of from water-level recording equipment at a properly-built cross-section
a measured parameter. There is often a conflict between the number of by means of a discharge rating curve determined from a number of
observations a measurement program can afford and the number discrete measurements taken by current meters and floats. Un-
needed to obtain an accurate and reliable parameter estimate. For this fortunately, this systematized measuring scheme is not available in all
reason, it is useful on the one side to investigate the propagation of river bodies world-wide. In this case, mobile measuring equipment is


Corresponding author.
E-mail address: ebekri@upatras.gr (E.S. Bekri).

https://doi.org/10.1016/j.jenvman.2019.05.064
Received 7 January 2019; Received in revised form 24 April 2019; Accepted 17 May 2019
0301-4797/ © 2019 Elsevier Ltd. All rights reserved.
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

employed. The most common and reliable method is to determine the measurement errors due to measurement noise. It is applied mainly in
geometric properties of the cross-section in conjunction with the flow industry sectors where the measurements are inaccurate or even non-
velocity, employing a current-meter at specific depths. However, other existent, for example: in the upstream sector where flow meters are
accurate river-discharge measuring or estimating techniques have been difficult or expensive to position (Delava et al., 1999); or where accu-
proposed; these include the Acoustic Doppler Current Profiler rate data is of high importance, for example for security reasons in
(Simpson, 2001; Gartner and Ganju, 2007), although this has proven nuclear power plants (Langenstein et al., 2004). The DVR improves the
problematic during high river flows because of high-suspended sedi- accuracy of the process data by adjusting the measured values so that
ment concentration occurrences (Yorke and Oberg, 2002). In addition, they satisfy the process constraints. In general, data reconciliation can
combinations of measuring and computational techniques have been be formulated by a weighted least squares (WLS) optimization problem
developed (Koelling, 2004; Nakagawa et al., 2007; Kuriqi and or maximum likelihood objective function; where the measurement
Ardiçlioglu, 2018). It is of note that some of these methods interfere errors are minimized with process model constraints. These data re-
with the natural water quality or are prohibitively expensive and conciliation techniques not only reconcile the raw measurements, but
complex to install and operate (Kuusisto, 1996). also estimate unmeasured process variables or model parameters, if
For the determination of river water quality characteristics, in-situ they are observable. Usually, the kind of constraints included are mass
field measurements of physical and chemical parameters (such as and energy-balance constraints, such as deterministic valid physical
temperature, conductivity or salinity, dissolved oxygen, pH, turbidity, laws and, in some cases, other inequality relations imposed by the
fluorescence, BOD5, nutrients and metals) take place in parallel with feasibility of the process operations. Empirical or other types of equa-
water sampling for laboratory analysis (Kuusisto, 1996). Standards for tions, involving many unmeasured parameters, are not recommended
water quality monitoring could be “any relevant CEN/ISO standards or to be used as constraints, since at best they are only known approxi-
such other national or international standards, which will ensure the mately. The reconciled estimates are expected to be more accurate than
provision of data of an equivalent scientific quality and comparability” the measurements themselves and, more importantly, to be also con-
(WFD, 2000). It is worth mentioning that the use of tracers as water sistent with the known relationships between the process variables as
quality signatures has been recognized as the most productive method defined by the constraints.
in hydrology for determining water budgets and stream-flow generation The most commonly-used criterion/objective function to select the
processes (Peters, 1994). According to Peters (1994), environmental optimal solution in data reconciliation is the WLS deviations of the
tracers (such as naturally-occurring isotopes [18O, D], solutes [Cl−, corrected discharges from their measured values. However, as analyzed
Br−, SO42−]) and other physical and chemical characteristics (tem- in Özyurt and Pike (2004), different objective functions besides the
perature, specific conductance and alkalinity) used to track the move- WLS can be used for data reconciliation. Classical approaches to data
ment of water have gained widespread acceptance. Moreover, tracers validation are usually solved by statistical methods, when an explicit
can assist in identifying the spatial and temporal movement of water characterization of the measurement error is available. The main dif-
flowing into a catchment. During the tracer monitoring period, it is ficulty with this statistical method is that the processes are not always
essential to ensure that no unusual climatic conditions are taking place. perfectly described and understood, and, thus, the measurement pre-
Additionally, the water sampling position should be carefully selected, cision cannot be precisely quantified. In many cases, the user has only
verifying good vertical mixing conditions (Elhadi et al., 1984). Kirchner an experimental knowledge, or a rough guess based on previous theo-
et al. (2000) analyzed a detailed time series of chloride in both rainfall retical and experimental knowledge of the value range of a parameter
and runoff from a headwater catchment, demonstrating that con- (upper and lower limit). The concept of expressing the measurement
taminants will initially be flushed rapidly but then low-level con- error as an interval is employed in the so-called parameter set estima-
tamination will be delivered to streams for a surprisingly long time, tion from bounded error data (Fogel and Huang, 1982; Milanese and
revealing that the tracer concentrations will retain a long chemical Belforte, 1982; Himmelblau, 1985; Milanese et al., 1996; Ragot and
memory of past inputs. Maquin, 2005). In this estimation, it is assumed that all types of errors
In many cases, the time available for completing river-flow rate and belong to a known set and that the measurement error is bounded. It
water-quality measurements at various cross-sections (covering the includes the determination of a set of constant parameter values (called
entire river and its tributaries), ensuring steady state flow conditions, is the feasible parameter set) which is compatible with all the available
significantly shorter than the time needed for in-situ measurements and observations, taking into account the error bounds and the model
sampling during the daytime. In such cases, where budgetary con- constraints (Maquin et al., 1991).
siderations restrict the implementation of monitoring programs, quick Based on nonlinear optimization, this study proposes an error-cor-
methods which are low in cost and reliability (such as floats, release of rection methodology for computing river discharge and pollutant/
air bubbles and the pendulum) (Yannopoulos, 1995; Yannopoulos et al., tracer concentration values (and subsequently pollution loads) of
2000, 2008) are often employed for river discharge measurements. higher accuracy and reliability compared with the initial measurements
These measurements should then be subject to error correction to in- (Yannopoulos and Bekri, 2010; Bekri et al., 2012, 2013). In order to
crease their reliability. achieve this, water volume conservation is combined with pollutant/
Various methods have been developed for estimating the un- tracer mass balance within river balance nodes. This method also en-
certainty in discharge measurements, including hydrographer estimates ables the determination of a non-measurable unknown latent discharge
(Rantz, 1982) and statistical error propagation techniques (Carter and at each river node. It resembles to data reconciliation techniques, since
Anderson, 1963; Herschy, 1971, 1985; ISO, 1979; WMO, 1980; Sauer it also aims to correct the raw measurements by formulating an opti-
and Meyer, 1992). Error correction techniques are generally very cost- mization problem based on the mass conservation principle. However,
effective and, therefore, efficient for data assimilation. A thorough re- to the best of our knowledge, the combination of water volume and
view of processes for estimating discharge-measurement errors was tracer mass conservation in a river network by measuring cross-sections
published by Dickinson (1967), updated by Pelletier (1988) and lately and balance nodes (as analyzed here) has not been developed and ap-
by McMillan et al. (2012). The uncertainty of individual river-discharge plied up to now to correct the measurements of flowrate and pollutant
measurements includes random and systematic errors in the cross-sec- concentration with the aim of computing more reliable pollution loads.
tional area, as analytically described in Sauer and Meyer (1992) and Le We test the technique on a virtual river for which the true values of
Coz (2012). flowrate and pollutant/tracer concentration are defined. Various mea-
The data validation and reconciliation (DVR) methodology (Kuehn surement sets are produced with Monte-Carlo simulations incorporating
and Davidson, 1961; Mah et al., 1976; Himmelblau, 1978; Maquin the effects of random and systematic errors. Next, we apply the method
et al., 1991) is an interesting scientific effort aimed at correcting to a real case, the Alfeios river (Greece), where quantitative and

372
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

qualitative measurement data are available at various cross-sections


covering the entire river length and its tributaries. The suggested
methodology can be applied not only to rivers without permanent
measurement stations, as presented in this paper, but also to rivers
composed of proper measurement positions/cross-sections, resulting in
balance nodes. The technique can also be employed in cases where the
measurements are taken with more accurate measuring methods than
the ones described in this paper, in order to further increase the accu-
racy and to estimate any possible non-measurable latent quantities.
Within this framework, the proposed methodology attempts to reduce
time, complexity, personnel and the cost of river monitoring programs,
as well as of water resource management plans.

2. Material and methods

2.1. Problem setup

Let us consider an ungauged river of interest which is composed of a


main river and some tributaries. Within a limited budget, we need to
apply a monitoring program for the water status of this river with the
aim of specifying changes in water quality and possible water pollution
as well as assessing remediation measures. A sufficient number of
measurement positions/cross-sections should be selected, covering the
entire river length and its most important tributaries, where parallel
measurements of water quantity and quality will take place. The loca-
tion of each measurement position must be correctly selected to ensure
that the cross-sections are situated close enough to minimize any in-
termediate water inflow. On the other hand, the cross-sections should
be located far enough away from each other to allow for complete cross-
sectional mixing conditions. It is then possible to define balance nodes
for which the water volume and mass conservation of various stable
and conservative pollutant/natural tracers can be estimated. Each
single node is composed of one inflowing and one outflowing cross- Fig. 1. (a) Schematization of an example river into 6 measurement cross-sec-
section situated on the main river; and the intermediate cross-sections, tions and 2 balance nodes (see Section 2.3 for more details); (b) General no-
if any, correspond to its tributaries. A common cross-section is defined tation of the proposed technique for two intermediate balance nodes i = k and
i = k+1 of a general example river and the corresponding cross-sections (i,s).
between successive nodes, which for the downstream node is inflow
and for the upstream node is outflow.
In each single-balance node, we account for the existence of an
unknown and not directly-measured water quantity (referred to as to the node's water volume balance and which may be taken into
“latent”) in the process. The latent discharge of a balance node is as- consideration by defining a latent term.
sumed to correspond to the runoff of a catchment area, which is in- The days of the measuring expeditions should be carefully selected,
cluded between all inflowing cross-sections and the outflowing cross- ensuring that the measurement conditions refer to the mean hydraulic
section of the node. Τhe runoff from this area is missing from the water conditions usually prevailing in the flow under consideration (Schmidt
balance of the node. The flow rate, which is measured at each inflowing et al., 2012), i.e. being in a steady state (no transient effects). In order to
cross-section of a node, is the sum of the drainage area of the corre- minimize the duration of each measurement expedition (one to two
sponding sub-catchment up to the given cross-section. On the other days), quick river-discharge measurement methods (which are low in
hand, the flow rate measured at the outflowing cross-section is the sum cost and reliability) together with in-situ field measurements of pollu-
of the entire area of the whole catchment up to the given cross-section, tants/natural tracers can be undertaken; thus, the overall cost of each
which is not considered from the inflowing cross-sections. However, the expedition will remain low. In order to estimate reliable pollution
exact area for the latent quantity cannot be computed with certainty loads, accurate river discharge and pollutant/tracer concentrations are
and only rough approximations may be made. In addition to con- required. For our study, we select pollutant/tracer concentrations
sidering the unaccounted areas, the latent discharge is assumed to in- which have sufficient levels of measurement accuracy. When pollutant
clude any other unknown or hidden contribution from the surface water or natural tracers are measured precisely, the accuracy of discharge
bodies and/or their interplay with the groundwater, which is expected measurements becomes the most critical component of the pollutant
to be quite small based on the assumptions for the application of the load computation, and the largest source of error (NCSU, 2008). Thus,
proposed methodology. Thus, according to our latent term definition, within our research framework, discharge measurements with low re-
the latent discharge may flow into or out of the corresponding node. It liability levels should be corrected to increase their accuracy.
should be noted that the necessity of considering a latent term is also To facilitate comprehension of the proposed methodology, a sim-
verified from the theory of data reconciliation (e.g. Ragot et al., 1991; plified example river (Fig. 1(a)) will be presented in parallel with the
Baloochy et al., 2010; Hancock, 2013). More precisely, in cases where general description. This example river is composed of the main river
significant losses are present, the inclusion of an unknown loss term in and three tributaries. Six measurement cross-sections are specified,
the balance equation, which can be estimated as part of the re- defining two balance nodes. These two successive nodes are then con-
conciliation equations, has been suggested. In river systems, even when nected with a common cross-section (cross-section 3). The methodolo-
the cross-sections have been properly selected as previously discussed, gy's overview is given in the flowchart of Fig. 2 and a glossary including
there are unknown and unobserved water quantities which contribute abbreviations and symbols is given in Appendix B.

373
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

corrected for all cross-sections; and (ii) the dual inequalities (in the
form of upper and lower value bounds) to be derived by properly re-
formulating the equations of water volume and pollutant mass con-
servation at each single-balance node, as well as at all possible balance-
node combinations (i.e. every two successive nodes, every three suc-
cessive nodes, up to the balance of the entire river).
The measurement errors for the river discharge are approximated
based on the author's experimental knowledge. From the total of the
measured pollutants/natural tracers, we select those which have been
estimated sufficiently accurately; therefore, resulting in a sufficiently
low measurement error. The measurement error of the pollutant/tracer
concentration is computed based on the measuring equipment error
parameters provided by the manufacturers.
The electrical conductivity of water (hereafter called “conductivity”
and symbolized with “EC”) is one of the most commonly-measured
physico-chemical parameters and, as such, is included in the applica-
tion of the proposed methodology. The increasing use of conductivity as
a natural tracer is related to the growing availability of commercial
sensors enabling highly reliable and simple measurements (Schmidt
et al., 2012). Conductivity is considered a good estimate of the total
inorganic dissolved solids present in the water column (Eaton et al.,
1995; Allan et al., 2007). The total dissolved solids (TDS) concentration
is calculated as the summation of anions and cations dissolved in water
(inorganic salts – mainly magnesium, calcium, sodium, potassium,
chlorides, sulphates and bicarbonates), and is considered to be an in-
direct measure of the water quality with respect to the amount of dis-
solved ions; since it does not provide analytical information regarding
the nature and the exact relationship of the present ions, nor the water
characteristic parameters. Thus, the TDS concentration plays the role of
a general water-quality indicator. The conductivity value is directly
proportional to the TDS concentration. It is therefore possible to use
conductivity measurements in the pollutant conservation equation in-
stead of the TDS concentration. The approximate conversion of water
conductivity (usually expressed in mS/cm) into TDS concentration
(expressed in ppm) is undertaken through a factor ranging from 0.5 to
0.9, depending on the chemical composition of the TDS (APHA, 1999).
Moreover, chloride and sulphate ions are also included in this process.
Kim et al. (2002) investigated the chemical behavior of major inorganic
ions in the streams of the Mankyung river area (South Korea). The
authors revealed that (i) chloride and sulphate concentrations, (ii) the
total concentration of major cations, and (iii) conductivity in the stream
were controlled by mixing, indicating a conservative behavior like
chloride. The same study demonstrated that the alkalinity and con-
centration of nitrates are regulated by various reactions such as mixing,
photosynthesis, respiration and decomposition of organic matter (ibid);
for this reason, we do not include them in our methodology.
In each single-balance node, we account for the existence of an
Fig. 2. Flowchart of the steps of the proposed error correction methodology.
unknown and not directly-measured latent water quantity in the pro-
cess, as described in Section 2.1. Concerning the unknown measure-
2.2. Conceptual framework ment/estimation error of the latent discharge terms, we consider a
larger relaxed value. The reason for this is to avoid the unmeasured
Based on the schematization of a river into properly-selected mea- latent terms taking values close to their initial estimates (since these are
surement cross-sections and balance nodes as described in Section 2.1, not measured) and the water volume and tracer mass conservation
we propose an original methodology for correcting mainly river-dis- balances from which they result being subject to errors. In this way,
charge measurements (these being the most critical parameter for the these hypothetical latent terms cannot play a divergent role in the
estimation of pollution loads). Pollutant/tracer concentrations can also optimization of the values of the measured cross-sections, since they do
be corrected with the suggested correction technique, though the pre- not restrict the balances into narrow value limits. This argument for the
cision of their measurement values must already be sufficiently high, in unmeasured variables is also proposed by Mandel et al. (1998) and
order to meet the requirements of the method. Ragot and Maquin (2005).
The method includes formulating a nonlinear optimization problem
based on the measurements and corresponding measurement errors of 2.3. Notation and symbols
the two parameters: a) river discharge and b) pollutant/tracer con-
centration at the defined cross-sections. The objective function is the To present the mathematical form of the proposed methodology, we
WLS deviations of the corrected discharges from their measured values. employ the general notation detailed below for the cross-sections and
The constraints include: (i) the dual inequalities (in the form of upper the nodes, and for two successive nodes, as shown in Fig. 1(a). This
and lower value bounds) of the two parameters to be optimized/ notation is accordingly extended for more balance nodes for every river

374
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

of interest, as shown in Fig. 1(b) for two successive balance nodes. The 2.4. Mathematical framework: nonlinear optimization
balance nodes are symbolized with i and their enumeration starts from
the river springs (i = 1) towards river estuaries (i = K), following the The chosen objective function, F, includes the WLS deviations of the
flow direction. The enumeration of the cross-sections also follows the corrected river discharges from their measured values for all cross-
flow direction. They are denoted by using two indices (i,s): the first one, sections, including the latent ones. The corrected values of the para-
i, is connected to the node enumeration (where: i = 1,2, …,K+1) and meters are derived from the minimization of the objective function F:
the second one, s, to the enumeration of the cross-sections within node i K ni 2 2 K 2
(where: s = 0,1,2, …,ni and ni is the total number of tributaries at node Xi, s x i, s XK + 1,0 xK + 1,0 Xi, x i,
F= + +
i). i=1 s=0 i, s K + 1,0 i=1 i,

More precisely, for all cross-sections situated along the main river (1)
(not corresponding to tributaries) the index s is set equal to 0 and the
The constraints of the optimization problem include two groups of
index i takes the value of the node into which the cross-sections flow.
equations. Firstly, dual inequalities express the upper and lower ac-
This results in the following indices: (1,0) denotes the inflowing cross-
ceptable value bound of the parameters to be optimized, based on their
section of node i = 1; (2,0) denotes the inflowing cross-section of node
measured values and the corresponding measurement error, as shown
i = 2,3,4 etc.; and (K,0) denotes the inflowing cross-section of the last
below:
node i = K. Following this notation, for the last defined node K the
outflowing cross-section is denoted (K+1,0). The node i = K+1 is a 0 xi, s (1 i, s ) Xi,s xi, s (1 + i, s ) for 1 i K + 1; 0 s ni; s = ;
hypothetical node situated downstream from the last defined node K 0 yi, s, j (1 i, s , j ) Yi, s, j yi,s, j (1 + i, s, j) for 1 i K + 1; 0 s ni; s = ; 1 j p

and is considered only in order to conform to the notation system. For (2)
illustrative reasons, this general notation is applied to the example river
The second group of constraints is based on the equations of water
shown in Fig. 1(a). In this case, the three cross-sections situated along
volume and pollutant/tracer mass conservation written for each single-
the main river are denoted as (1,0), (2,0) and (3,0).
balance node, as well as for every possible combination of the succes-
The measurement cross-sections (i,s) situated on the tributaries are
sive balance nodes (every two balance nodes, every three balance
defined by employing as i the number of the node to which they belong
nodes, up to taking the balance of the entire river). The group includes
(where: i = 1,2, … K), and as s the number from 1 up to the number of
(a) equality and (b) dual inequality constraints as analyzed below. The
tributaries of node ni following the flow direction. For the example river
equality constraints are the balance equations. The water volume and
in Fig. 1(a), the first node i = 1 is composed of one tributary and the
the pollutant/tracer mass conservation equations written for a combi-
corresponding cross-section is denoted as (1,1); and the second node
nation of successive nodes (more precisely, from node i = a up to node
i = 2 comprises two tributaries and the cross-sections (2,1) and (2,2).
i = b) are given as follows:
The latent term (which is not measured but estimated as described in
Section 2.1) is signified by the Greek index s = λ. For the example river DX b =X ,0 +(
b ni
X ± i = Xi,
b
) Xb + 1,0 for 1 a b K
i= s = 1 i, s
in Fig. 1(a), there are two cross-sections corresponding to the latent b ni b
Dq b,j = X ,0 Y +( X Y ± Xi, Yi, ,j)
water quantities of the two nodes, those being (1,λ) and (2,λ). ,0, j i= s = 1 i, s i, s, j i=

The measured values of the river discharge and of the concentration Xb + 1,0 Yb + 1,0, j for 1 a b K; 1 j p
of a pollutant/tracer j at a cross-section (i,s), as well as the corre-
(3)
sponding estimated latent terms, are xi,s and yi,s,j (i = 1,2, …,K+1;
s = 0,1, …,ni, λ and j = 1,…,p). It is assumed that the total number of By setting a = b and considering the residuals DXab and Dqab,j of the
appropriate pollutants/tracers with measured concentration is p balance equations equal to zero, expressions (3) can be written for the
(where: 1 ≤ j ≤ p). The pollution load of a pollutant j at a cross-section balance (water volume or pollutant mass balance) of each single node i
(i,s) is denoted as qi,s,j, and is estimated as the product of the measured (where: i = 1,2 …,K).
values of the water flow rate and the pollutant concentration: xi,syi,s,j. A first rough estimation of the unknown values of the latent terms
The corrected/optimized value (derived from the proposed metho- xi,λ and yi,λ,j (which cannot be directly measured) can be computed
dology) of the river discharge and of the pollutant concentration are from expressions (3) and written for each single node by using the
denoted as Xi,s and Yi,s,j. The absolute maximum relative measurement measured values xi,s and yi,s,j, instead of the corrected ones Xi,s and Yi,s,j.
error of the river discharge at a cross-section (i,s) is denoted as εi,s, and The latent term xa,λ of node i = a is then derived as follows:
the absolute maximum relative measurement error of the concentration na

of a pollutant j is written as ζi,s,j. In Table 1, we present an overview of x a, = DXaa x a,0 xa, s + Xa + 1,0 for Dxaa = 0
(4)
the notation and the input and output data of the proposed error-cor- s=1

rection technique for the example river shown in Fig. 1(a). Next, a first estimate of the unknown value yi,λ,j of the latent pol-
lutant/tracer concentration of node i and for each considered pollutant j
is provided by the pollutant mass conservation of the node. For node

Table 1
Notation and symbols of river discharge and tracer concentrations at each cross-section of the river in Fig. 1(a): a) measured/estimated and b) optimized/corrected
values.
Cross-section Node River discharge Concentration for each pollutant/tracer j

a/a Notation Measured Optimized Measurement error Measured Optimized Measurement error

1 (1,0) 1 x1,0 X1,0 ε1,0 y1,0,j Y1,0,j ζ1,0,j


2 (1,1) 1 x1,1 X1,1 ε1,1 y1,1,j Y1,1,j ζ1,1,j
Latent1 (1,λ) 1 x1,λ X1,λ ε1,λ y1,λ,j Y1,λ,j ζ1,λ,j
3 (2,0) 1&2 x2,0 X2,0 ε2,0 y2,0,j Y2,0,j ζ2,0,j
4 (2,1) 2 x2,1 X2,1 ε2,1 y2,1,j Y2,1,j ζ2,1,j
5 (2,2) 2 x2,2 X2,2 ε2,2 y2,2,j Y2,2,j ζ2,2,j
Latent2 (2,λ) 2 x2,λ X1,λ ε2,λ y2,λ,j Y2,λ,j ζ2,λ,j
6 (3,0) 2 x3,0 X3,0 ε3,0 y3,0,j Y3,0,j ζ3,0,j

375
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

i = α, the ya,λ,j is computed as follows: Next, it uses the branch-and-bound technique to exhaustively search
na these sub-problems for a global solution.
(Dqaa,j x a,0 ya,0, j x y
s = 1 a, s a,0, j
+ x a + 1,0 ya+ 1,0, j )
ya, =± for Dq =0
(5)
,j ,j
x a,
3. Results and discussion
The inequality constraints of the second group are formulated by
replacing in expressions (2) the Xi,s and Yi,s,j for each cross-section (i,s) 3.1. Virtual river example
(where i = 1,2 …,K and s = 0,1 …,ni,λ) with its equivalent as derived
from the balance equality (3). For example, regarding cross-section (b To test the consistency of the proposed methodology, we considered
+1,0), based on the water volume balance from node a up to b (ex- a “virtual example” of a river, as shown in Fig. 1. More precisely, for
pression (3)), the following dual constraint can be formulated: this virtual example we used a sufficiently large number of simulated
measurement datasets employing the Monte Carlo technique. For real
b ni b
0 xb+ 1,0 (1 b + 1,0 ) DX b + (X ,0 +( i=
X
s = 1 i, s
± i=
Xi, )) rivers, it was not possible to find complete and detailed datasets, as
xb+ 1,0 (1 + b + 1,0 ) required for fully testing the methodology's consistency. The reason for
(6) this is that the required datasets should include extensive quantitative
and qualitative measurements with both accurate (representing the real
For cross-section (b+1,0), the inequality constraint of this form will values) and inaccurate (representing the measured values subject to
be written as many times as the number of balance equations of every correction) methods employed at numerous measurement cross-sec-
single- and multiple-node combination, which includes the given cross- tions. The measuring cross-sections should cover the whole river length
section. (all tributaries and the main river) and should be properly situated to
For the pollutant/tracer mass conservation, the corresponding in- enable the formulation of balance nodes. Moreover, the measurement
equalities take the following form for cross-section (b+1,0), based on datasets should include also accurate measurement of the latent
the balance from node a up to b (expression (3)): quantities (which is practically impossible since these are unobserved
quantities) in order to enable the comparison of the corrected (from the
0 xb+ 1,0 (1 b + 1,0 ) yb + 1,0, j (1 b + 1,0, j )
methodology) latent quantity with the real values represented from the
b ni b
( Dqab + (Xa,0 Ya, o, j + ( i=a
X Y
s = 1 i, s i, s, j
± i=a
Xi, Yi, , j ))) accurate measurements. In real rivers, such detailed sets of measure-
xb+ 1,0 (1 + b + 1,0 ) yb + 1,0, j (1 + b + 1,0, j ) (7) ments are not available and from economical point of view not possibly
collected. After testing the consistency of the methodology, to demon-
Appendix A details the optimization problem for the example river strate its application at any river with measuring cross-sections for-
in Fig. 1(a). mulating balance nodes, a real case study, being the Alfeios river in
Greece, is considered in Section 3.3.
2.5. Computer implementation The “virtual river example” was built as follows. Accurate mea-
surement values of river discharge and pollutant/tracer concentrations
To implement the proposed methodology, we selected the LINGO were theoretically determined by the authors in order to fully satisfy
optimization software (Lindo Systems Inc., 1996; Schrage, 1997), since the water volume and pollutant mass conservation requirements. These
it is a very efficient and robust tool for building and solving mathe- are given in Table 2 as the real values of the two parameters (river
matical optimization models. To increase the flexibility and ease of the discharge and tracer concentration), denoted as Xreali,s and Yreali,s,j. In
proposed methodology, Microsoft Excel was also employed in order to order to produce a sufficiently large number of measurement sets of xi,s
import and export input and output data to and from LINGO. This was and yi,s,j for all river cross-sections, Monte-Carlo simulations were un-
made possible through the OLE Automation Links from Excel. Most of dertaken, as described below. The wide range of the produced mea-
the input computational processes necessary for the calculation of a) surement sets represents various random measurements undertaken at
the coefficients of the objective function and constraints, and b) the different random moments at the considered cross-sections across the
right- and left-hand side of the constraints of the optimization algo- river. Moreover, the effect of systematic errors (SE) was also included in
rithm, were introduced into Excel through VBA macros, which com- order to consider a problem of the measurement data handling system,
municated with LINGO to exchange data and run the algorithm at every or a misuse of the measuring instrument. For the SE we assume low,
step. To compute the optimization problem of our proposed metho- medium and high values (Table 3) with positive and negative sign to
dology, we employed LINGO's Global solver. Local search solvers are cover a systematic under- and over-estimation of the real values. Based
generally designed to search only until they have identified a local on these, we produced simulated measurement sets considering the
optimum. However, the Global LINGO solver converts the original non- following eight combined cases of random errors (RE) and systematic
convex, nonlinear problem into several convex, linear sub-problems. errors (SE):

Table 2
Real values and measurement errors of river discharge and tracer concentrations at each cross-section of the river in Fig. 1(a).
Cross-section Node Real river discharge, Xreali,s (m3/s) Measurement error of river discharge, εi,s Real concentration, Yreali,s,j, for Measurement error, ζi,s,j, for

a/a Notation EC (mS/cm) SO42−(g/l) EC SO42-

1 (1,0) 1 7 1.00 0.60 0.100 0.10 0.15


2 (1,1) 1 10 0.35 0.39 0.040 0.10 0.15
Latent1 (1,λ) 1 3 0.90 0.43 0.033 0.30 0.30
3 (2,0) 1 and 2 20 0.05 0.47 0.060 0.10 0.15
4 (2,1) 2 37 0.06 0.43 0.017 0.10 0.15
5 (2,2) 2 7 0.05 0.32 0.006 0.10 0.15
Latent2 (2,λ) 2 2 0.90 0.42 0.055 0.30 0.30
6 (3,0) 2 66 0.05 0.43 0.030 0.10 0.15

376
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

Table 3
Low, medium and high SE values for the example river in Fig. 1(a): values expressed as a) % of the Xreali,s and b) as absolute.
Cross-section (1,0) (1,1) (1,λ) (2,0) (2,1) (2,2) (2,λ) (3,0)

3
Xreali,s (m /s) 7 10 3 20 37 7 2 66
εi,s 1 0.35 0.9 0.05 0.06 0.05 0.9 0.05
Low SE value SE (%Xreali,s) 5 5 5 1 1 1 10 1
SE 0.35 0.5 0.15 0.2 0.37 0.07 0.2 0.66
Medium SE value SE (%Xreali,s) 10 10 10 2 2 2 10 2
SE 0.70 1.00 0.30 0.40 0.74 0.14 0.20 1.32
High SE value SE (%Xreali,s) 20 15 20 3 3 3 20 3
SE 1.4 1.5 0.6 0.6 1.11 0.21 0.4 1.98

Case 1. RE following the uniform distribution (UD) and no SE (called cross-sections except for the latent ones; (ii) the OV standard deviations
UDN) are significantly lower in comparison with the MV ones. For the second
category: (i) the OV and MV averages are approximately equal (relative
Case 2. RE following the normal distribution (ND) and no SE (called
difference ≤0.05%); (ii) the OV averages are close to the real values,
NDN)
but not necessarily closer than the MV averages; (iii) the OV standard
Case 3. RE following the ND and low positive values of SE (called deviations differentiate slightly in comparison with the MV standard
NDL+) deviations. The small differences between OV and MV in this case can
be explained, since the river-discharge measurement errors are low and
Case 4. RE following the ND and low negative values of SE (called
the value range narrow. Concluding, for both categories, the proposed
NDL−)
methodology achieves to reduce the value range of the river discharge
Case 5. RE following the ND and medium positive values of SE (called measurements of all measured cross-sections, and the OV averages are
NDM+) close to the true value.
Case 6. RE following the ND and medium negative values of SE (called
NDM−) 3.2.2. Deviation of the MV and OV averages from the real values for the
latent cross-sections
Case 7. RE following the ND and high positive values of SE (called Regarding Table S1, the results for the latent quantities can be
NDH+) presented in three groups:
Case 8. RE following the ND and high negative values of SE (NDH−)
Based on the defined measurement errors εi,s and ζi,s,j (Table 2), 1) For the NDN case, the OV averages are almost equal to the corre-
1000 measurement error values (denoted as MCExi,s for discharges and sponding MV averages, and both are very close to the real value.
MCEyi,s,j for concentrations) were generated by Monte-Carlo simula- Therefore, the OV averages seem to be a very good estimation of the
tions for ND, assuming: a) N(0,εi,s/5) for the river discharge measure- theoretically non-measurable latent discharges. The OV standard
ment error, and b) N(0,ζi,s,j/5) for each pollutant/tracer concentration deviations are higher but comparable with the MV standard devia-
(to approximate almost 100% of the ND curve, we take εi,s = 5σ and tions.
ζi,s,j = 5σ). For the UD, the MCExi,s and MCEyi,s,j were generated based 2) For the cases with SE (being cases from 3 to 8), the OV averages start
on the minimum and maximum allowable values of each parameter, as diverging progressively from the MV averages and the real values as
shown in expressions (2). The 1000 measurement sets (xi,s and yi,s,j) the SE values increase from low to high. The OV standard deviations
were then computed as follows: xi,s = Xreali,s(1+ MCExi,s) and are almost the same as the MV standard deviations.
yi,s,j = Yreali,s,j(1+ MCEyi,s,j). In the above-mentioned error cases the 3) For the UDN case, the OV averages diverge from the MV averages
effect of the systematic errors (SE) on the process is considered only in and the real values, being comparable to the corresponding values of
combination with the random errors following the ND, since random the NDH+ and NDH− cases. The OV standard deviations are the
measurement errors often have a Gaussian normal distribution (Sauer highest, being two to three times higher than the standard devia-
and Meyer, 1992; Le Coz, 2012). The UD is also considered in order to tions of the measurement sets.
test the efficiency of the proposed methodology for an extreme dis-
tribution. 3.2.3. Convergence of the correction methodology towards the true value
Additionally, we tested if, and to what extent, the proposed cor-
rection method converges towards the true values. To this end, we
3.2. Results and discussion of the virtual river example
plotted the diagrams (Fig. 3) for each cross-section, as well as for the
sum of the differences of all cross-sections together. The x-axis is the
The averages and the standard deviations of (a) the 1000 mea-
absolute difference between the MV and the real discharge value, │xi,s-
surement sets (measurement values - MV) generated by Monte-Carlo Xreali,s│; and the y-axis is the absolute difference between the OV and
simulations and (b) their resulting optimized values (OV) for all error the real discharge value, │Xi,s-Xreali,s│. The ideal correction by the
cases are presented in the supplementary material (Table S1). The proposed methodology is the one for which the point of this diagram
presented results can be grouped into two categories based on the river- approaches or lies on the x-axis, where │Xi,s-Xreali,s│ = 0. The 45o-line
discharge measurement error of each cross-section: (a) cross-sections of this diagram (red solid line) represents the line where the MV and OV
with medium to high river-discharge measurement errors εi,s > 0.1 have the same distance from the real value. A percentage of ± 10%
and (b) cross-sections with low river-discharge measurement errors around the 45o-line is considered an acceptable tolerance of the dif-
(εi,s ≤ 0.1). ference │Xi,s-Xreali,s│. This 45o-line ± 10% tolerance (green dotted
line) is taken as the limit below which the measurements are either
3.2.1. Deviation of the MV and OV averages from the real values for the corrected toward the real value or their correction remains redundant.
measured cross-sections As a measure of good correction of the proposed methodology, the
Based on Table S1, for the first category: (i) for all error cases the percentage (%) of the number of the measurement sets situated below
OV averages are closer to the real values than the MV averages for all the 45o-line+10% (from now on called PE), which is the acceptable

377
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

Fig. 3. Absolute differences of MV (x-axis) and of OV (y-axis) from the real river discharge for: (a) the NDN case considering all cross-sections together, (b) the NDN
case for cross-section (1,0); (c) the NDN case for cross-section (1,2); and (d) for the NDH− case of cross-section (1,0).

correction area, was computed and is shown in the supplementary 3.2.4. Effect of the measurement error value on the convergence towards the
material (Table S2) for each cross-section and for the total sum. For the true value
NDN case, the corresponding diagrams demonstrate: i) the sum of all Finally, in all diagrams in Fig. 3 a lower limit is evident (marked
cross-sections together in Fig. 3(a); ii) cross-section (1,0) with the with a red closed frame in Fig. 3(d), where this limit is highlighted more
highest εi,s in Fig. 3(b); and iii) cross-section (2,1) with the lowest εi,s in clearly). This border shows that something is preventing the correction
Fig. 3(c). Generally, for all diagrams, even when the │xi,s-Xreali,s│ is algorithm from correcting these measurement sets, resulting in the
high (> 2), the │Xi,s-Xreali,s│ lies within the acceptable correction area. optimized values not being close to the ideal solution. To investigate
Focusing on Table S2, the following remarks can be made: i) Con- this, we analyzed the points of the limit in Fig. 3(d). We realized that
sidering all cross-sections together, the PE is very high (92–99%), re- they correspond to the lowest values of the generated measurement
vealing a satisfactory result for the method in general. ii) For cross- sets. Their allowable upper limit, x1,0(1+ε1,0), does not include the real
section (1,0) with the highest εi,s, the PE is relatively lower (73–80%) value and is lower, resulting in │Xi,s-Xreali,s│ > 0. For this reason, the
than the PE for the sum of all cross-sections together. iii) For the rest of correction technique computed their optimized values equal to their
the measured cross-sections, the PE is high, ranging from 88% to 99% allowable maximum value. Let us select a point from this limitation in
for all NDN cases, and is wider for the UDN case (64–93%). iv) For the Fig. 3(d) which corresponds to a measured value x1,0 = 1.69 m3/s,
latent quantities, the PE is significantly lower (under 50%) than for the where: (ε1,0 = 1.0; x1,0(1+ε1,0) = 3.34 m3/s; X1,0 = 3.34 m3/s;
measured cross-sections. v) Through a detailed analysis of the mea- Xreal1,0 = 7 m3/s). We considered various values of the corresponding
surement sets that lie in the unacceptable area of correction, it is of note measurement error ε1,0 (ε1,0 = 1,2,3,4,5,6,7,8,9,12,20,50,100) and, in
that the relative error (Xi,s-Xreali,s)/Xreali,s is not higher than 35% in Fig. 4(b), we computed and plotted: a) the allowable maximum value
most cases. More than 80% of the measurement sets in this area have limit (blue solid line) and b) the optimized values X1,0 (red dotted line).
absolute values less than ± 20% of the Xreali,s. This can be clearly seen From this figure, it is evident that: (a) for the values of ε1,0, where the
for cross-section (1,0) of the NDH− case, with relative error (2–37%). allowable maximum value limit is lower than the real value, a linear
The ordinary and cumulative histogram is presented in Fig. 4(a), line (first part of the red line) is created, and in this region the opti-
showing that more than 95% of the cases that violated the acceptable mized values are equal to this upper value limit; and (b) for the values
limit of correction have a relative error below 25%. of ε1,0, where the allowable maximum value limit is higher than the real

378
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

Fig. 4. For cross-section (1,0) and for the NDH− case: (a) ordinary and cumulative histogram of the relative error, (X1,0-Xreal1,0)/Xreal1,0, considering the MV
situated within the unacceptable area of correction; and (b) change of the optimized river discharge values X1 by increasing the measurement error ε1,0, together with
the upper value limit of the river discharge x1,0(1+ε1,0).

value, the optimized value of the river discharge is stabilized and re-
mains equal to X1,0 = 8.14 m3/s, regardless of the increase in the ε1,0
(horizontal part of the red line). It is worth noting that the increase in
the ε1,0 does not affect the optimized values of the river discharge in the
other cross-sections of the river.

3.3. Application on a real river basin

Next, we applied our methodology to a real river, the Alfeios river in


Peloponnisos, Greece (Manariotis and Yannopoulos, 2004; Bekri and
Yannopoulos, 2012; Podimata and Yannopoulos, 2013). The simulta-
neous discharge measurements (using quick techniques and water
sampling) included twelve cross-sections along the main river and its
tributaries, as shown in Fig. 5. The measurement cross-sections com-
prised either road bridges or dams, where access to the entire river
length was possible. Four measurement expeditions took place in 2006
and 2007 within the framework of the research program Pythagoras II-
Environment (Yannopoulos et al., 2007; Yannopoulos, 2008). The dis-
charge measurements included an estimation of the cross-section geo-
metrical characteristics along with the surface maximum velocity; this
was achieved by employing the floating-object method, the pendulum
method, and/or the air-bubble release method, depending on whether
the flow depth was irregular or whether the flow permitted bubbles to
be clearly viewed. The duration of the measurement expedition was
about one day in order to avoid or reduce the effect of a possible change
in the steady-state flow conditions.
Additionally, the date of each expedition was carefully selected
based on the weather data available (excluding dates directly after
short or long rainy periods); thus, every attempt was made to measure
the mean, steady-state flow conditions without bias. The travel time of
water from the river springs to the river delta was estimated to be about
2 days. Every expedition started with measurements at the measure-
ment cross-section closest to the estuaries of the Alfeios river, and then
moved upstream to the very last cross-section closest to the river
springs. For the sake of brevity, in this paper we present only the Fig. 5. Geographical depiction of the Alfeios river basin with the measurement
measurement set of the first measurement expedition, which took place cross-sections and corresponding sub-basins.
on 9 April 2006, from 07:40 till 20:00; this is because the application of
our correction methodology to all the measurement expeditions re- tributary, the Lousios river cross-section (1,1). The next downstream
sulted in similar conclusions. adjoining node i = 2 is composed of cross-sections (2,0) (common with
For the application of our technique, four balance nodes were the node i = 1) and (3,0) on the main Alfeios river, and contains two
carefully defined in order to satisfy the previously-discussed distance tributaries, the Ladhon river cross-section (2,1) (which covers almost
requirements of the methodology, covering the entire river length and one third of the total catchment) and the Erymanthos river cross-section
its tributaries. Starting from the river sources, the first node i = 1 in- (2,2). The third node i = 3 includes cross-section (4,0), being common
cludes cross-sections (1,0) and (2,0) on the main Alfeios river and one with the previous node, and cross-section (3,0), situated at the Flokas

379
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

Table 4
River discharge and tracer concentration at each measured (thus excluding the latent terms) cross-section of the Alfeios river: a) Measurements, b) optimized values
and c) measurement errors.
Cross-sections

(1,0) (1,1) (2,0) (2,1) (2,2) (3,0) (3,1) (3,2) (3,3) (4,0) (4,1) (5,0)

xi,s (m3/s)
3.01 6.68 19.66 42.00 7.08 67.70 0.37 0.32 2.45 67.00 1.54 66.50
Xi,s (m3/s)
6.02 10.02 19.04 40.08 7.01 67.69 0.38 0.34 2.51 66.04 1.46 67.51
εi,s
1.00 0.50 0.05 0.30 0.05 0.05 0.05 0.05 0.05 0.05 0.05 0.05
yi,s,j, where j = EC (mS/cm) and where ζi,s,j = 0.10
0.62 0.38 0.45 0.41 0.31 0.42 0.70 1.08 0.42 0.42 0.50 0.42
Yi,s,j, where j = EC (mS/cm)
0.62 0.41 0.48 0.45 0.34 0.45 0.63 0.97 0.46 0.45 0.45 0.45
yi,s,j, where j = SO42− (g/l) and where ζi,s,j = 0.15
0.12 0.05 0.06 0.02 0.01 0.03 0.08 0.16 0.04 0.04 0.05 0.04
Yi,s,j, where j = SO42− (g/l)
0.12 0.05 0.07 0.02 0.01 0.03 0.10 0.18 0.03 0.03 0.05 0.03

Diversion Dam used for irrigation purposes. This third node i = 3 in- node. Therefore, they are taken as being equal to the concentration
cludes the following tributaries: a) Two tributaries with a low to minor values of the outflowing cross-sections of the node of each latent term
contribution to the node's water volume balance, i.e. Kladheos river (see Table 5). In this case, again, the measurement errors are set to high
cross-section (3,1) close to Ancient Olympia, and Selinous river cross- values (ζi,λ,j = 1, where: j = EC; and ζi,λ,j = 1.5, where: j = SO42−).
section (3,2); and b) the entrance to the irrigation canal close to the
Flokas Dam, being the outflowing cross-section (3,3). The fourth and 3.4. Results and discussion of the real river basin
last node i = 4 is located near the river's estuary (Kyparissiakos Gulf). It
encompasses cross-sections (4,0) and (5,0) on the main Alfeios river, The correction methodology employed in this paper generated the
and one tributary, i.e. the Enipeus river cross-section (4,1). optimized values for all measured cross-sections (presented in Table 4)
The following natural tracers have been tested and selected as the and for the latent cross-sections (in Table 5). Since there are accurate
most appropriate based on the requirements of the methodology tested measurements available for only few cross-sections of the Alfeios river,
here (Ziabras and Tasias, 1992): (a) Water conductivity, with mea- the verification of the corrected (optimized) values is possible indirectly
surements in the field corresponding to a satisfactory accuracy level based on expert judgment. The following conclusions can be made:
(ζi,s,1≤0.10); and (b) Sulphate ion concentration (SO42−), with mea- For cross-section (1,0), with the highest discharge measurement
surements derived from laboratory analysis of water samples, again error, the optimized value is computed as two times higher than the
corresponding to a satisfactory accuracy level (ζi,s,2≤0.15). measurement, revealing a significant underestimation of this mea-
Table 4 demonstrates the measured values of: 1) the river discharge surement. This cross-section can only be accessed with difficulty and is
xi,s (m3/s); 2) the electroconductivity yi,s,j, where: j = EC (mS/cm); 3) quite abnormal and irregular. The river bed has a considerable incline,
the concentration of sulphate ions yi,s,j, where: j = SO42− (g/l); and 4) with large extruding rocks which the water passes over, making it
the corresponding measurement errors εi,s and ζi,s,j. For the latent cross- difficult to measure this unobserved quantity. This anomaly contributes
sections (1,λ), (2,λ), (3,λ) and (4,λ) there are no measurements avail- to increased measurement errors combined with very low accuracy and
able, since by definition it is not possible to measure the latent terms. In explains the substantial underestimation of the measured values. In this
order to approximate the “measured” values of the latent terms for the case, our correction algorithm succeeds in moving the optimized value
river discharge xi,λ (which is necessary for our methodology), we rely towards the direction of the expected true value, which is higher than
on authors’ expert knowledge regarding the latent quantities and spe- the measurement. The next measurement cross-section (1,1) presents a
cific hydrogeological characteristics of the corresponding sub-basin slightly better picture, however it also has an irregular shape, a notable
areas. More precisely, the “unobserved” values of the latent river dis- incline and large rocks are present, thus resulting in medium-level
charge xi,λ are assumed to take the values of 1 m3/s and 1.5 m3/s re- measurement errors. Also, in this case the optimized value is higher
spectively, as presented in Table 5. The measurement errors for all la- than the measurement, as could be expected based on the profile of the
tent cross-sections are set to high values (εi,λ = 1) so that the wide value measurement cross-section. Contrarily, measurement position (2,0) is
range does not restrict the optimized values close to the assumed characterized by a gentle, bright and easily-accessible cross-section.
measured values, as thoroughly explained in the description of the The air-bubble release method was employed here to estimate the river
methodology. To approximate the “measured” values of the latent discharge with sufficient accuracy (with measurement error
terms for the two tracer concentrations under consideration, it is as- εi,s < 20%). Our correction technique computed an optimized value
sumed that these latent concentrations are mainly the product of the very close to the measurement, indicating a result consistent with the
mixing of all pollutant loads from the incoming contributions of the observed measurement conditions.

Table 5
Latent terms of river discharge and tracer concentration of the Alfeios river: a) assumed “measured” values, b) optimized values and c) “measurement” errors.
Cross-sections xi,s (m3/s), εi,s = 1.00 Xi,s yi,λ,j where j = EC (mS/cm), Yi,λ,j where j = EC yi,s,j where j = SO42− (g/l), Yi,λ,j where j = SO42−(g/l)
(m3/s) ζi,s,j = 1.0 (mS/cm) ζi,s,j = 1,50

(1,λ) 1.00 3.00 0.42 0.42 0.04 0.02


(2,λ) 1.00 1.55 0.42 0.58 0.04 0.08
(3,λ) 1.50 0.15 0.42 0.84 0.03 0.09
(4,λ) 1.50 0.00 0.45 0.82 0.06 0.10

380
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

For measurement cross-section (2,1) at the Ladhon river, there is a of their estimation (even if relatively inaccurate) is still very important
registered water volume value (provided by the Hellenic Public Power and useful.
Corporation) for the volume released by the Ladhon Hydropower sta-
tion situated at the most upstream point of the sub-basin (=36.75 m3/ 4. Conclusions
s). The hydrologic characteristics of the drainage area of this sub-basin
justify a small additional contribution from the surface runoff of the In this paper we developed an original error correction technique,
area downstream from the hydropower station (with a value of ≤3 m3/ enabling the estimation of more reliable river-discharge values mainly
s). The computed optimized value (40 m3/s) is lower compared with the in rivers without permanent measurement stations, compared with the
measurement (42 m3/s), and the optimization algorithm succeeds in measurements taken. It included a nonlinear optimization model based
moving towards the direction of the registered value. For measurement on the principles of water volume and pollutant mass conservation
position (2,2) at the Erymanthos river, our correction technique com- within single- and multiple-balance river nodes. In addition, the
puted an optimized value very close to the measurement, indicating a method considered probable, but unknown, latent quantities for each
result consistent with the observable measurement conditions. balance node. The proposed methodology can be applied to any river
For measurement cross-sections (3,0), (4,0) and (5,0) situated at which is composed of proper measurement positions/cross-sections
specific points based on the hydrogeological profile of the Alfeios river, resulting in balance nodes, as required by the technique. We verified
the river discharges are all of the same order without significant var- the consistency of our correction methodology through a virtual river
iations. This is in accordance with the optimized river-discharge values example, defining the real values of flowrate and pollutant/tracer
computed using our technique. For the low-flow cross-sections (3,1) concentrations and generating measurement sets based on Monte-Carlo
and (3,2), no significant variation of the optimized values from the simulations, including random and systematic errors. Our methodology
measured ones was generated by the algorithm. succeeded in achieving a reduction in the value range of the river
For measurement position (4,1), our correction technique computed flowrate compared with the range of the measurements. For the latent
an optimized value very close to the measurement, which is in ac- quantities, the average of the optimized values was close to the average
cordance with the expected results since the measurement conditions of the measurements; however, the average of the optimized values
were very good at this cross-section, resulting in a high level of mea- diverges from the average of the real values progressively, as the value
suring accuracy. of the systematic error increases. The methodology was also success-
The optimized value of the latent river discharge of the first node fully implemented on a real river basin. Therefore, it could constitute a
X1,λ (Table 5) has the highest value of all the latent values, and this is simple, efficient and valuable tool which could be implemented as part
consistent with the latent area of this node, which is the largest com- of catchment pollution monitoring programs.
pared to the other nodes. Moreover, in this latent area there is a
probable contribution of groundwater. The optimized values of the la- Acknowledgments
tent river discharges of the second, third and fourth nodes are also
consistent with the latent areas of the corresponding nodes. To test the This research has been co-financed by the European Union
effect of the assumed “measured” values of the latent discharges for all (European Social Fund - ESF) and Greek national funds through the
nodes, various values xi,λ (0.5,1,5, 2 and 3 m3/s) were considered. The Operational Program “Education and Lifelong Learning” of the National
optimized river discharge values of all cross-sections were not affected Strategic Reference Framework (NSRF) - Research Funding Program:
by these changes, showing a stabilized solution. Heracleitus II: Investing in knowledge society through the European
The corrected (optimized) values of the pollutant/tracer con- Social Fund. We would like to express our gratitude to: 1) Dimitris
centrations of all measured cross-sections are very close to the mea- Demetracopoulos, Ioannis Argyrakis, Ioannis Mavros and Ioannis
sured ones, with narrow value ranges; since, according to the metho- Stathas from the Hellenic Public Power Corporation for providing in-
dology proposed in this paper, the measurement errors should be very valuable operational data for the Ladhon hydropower station; 2) to
low (high measuring accuracy) in order to be included in the optimi- HYDROCRITES University Network for its support (http://www.
zation problem. The optimized values of the latent tracer concentra- hydrocrites.upatras.gr); 3) to Professor Markus Disse from the Chair
tions of all nodes, for which the measurement errors may have large of Hydrology and River Basin Management of the Technical University
values, cannot be explicitly evaluated since no measurements are Munich, Germany, for his constructive comments; and 4) to Professor
available. Despite this, it is possible to determine that these optimized Anastasios Stamou from the Department of Water Resources and
values are within the value range of the measured tracer concentration Environmental Engineering of the National Technical University of
of the Alfeios river basin. Generally, since the direct measurement of Athens for his valuable time and his constructive comments concerning
the latent terms is impossible, the estimation and subsequent correction the qualitative analysis of the river discharge measurements.

Appendix C. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jenvman.2019.05.064.

Appendix A. Application of the mathematical framework to the example river.

For the river in Fig. 1(a), the first type of inequalities is written as follows:
0 x1,0 (1 1,0 ) X1,0 x1,0 (1 + 1,0 )
0 x1,1 (1 1,1) X1,1 x1,1 (1 + 1,1)
0 x1, (1 1, ) X1, x1, (1 + 1, )
0 x2,0 (1 2,0 ) X2,0 x2,0 (1 + 2,0 )
0 x2,1 (1 2,1) X2,1 x2,1 (1 + 2,1)
0 x2,2 (1 2,2 ) X2,2 x2,2 (1 + 2,2 )
0 x2, (1 2, ) X2, x2, (1 + 2, )
0 x3,0 (1 3,0 ) X3,0 x3,0 (1 + 3,0 )

381
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

For each cross-section (i,s), i = 1,2 and s = 0,1,2,λ; and for each selected pollutant/tracers (j), 1 ≤ j ≤ p. Thus, the corresponding inequalities
based on the pollutant/tracer mass balance can be written as shown for the cross-section (1,0):
0 y1,0,j (1 1,0,j ) Y1,0,j y1,0,j (1 + 1,0,j )
0 y1,1,j (1 1,1,j ) Y1,1,j y1,1,j (1 + 1,1,j )
0 y1, , j (1 1, , j ) Y1, ,j y1, , j (1 + 1, , j )
0 y2,0,j (1 2,0,j ) Y2,0,j y2,0,j (1 + 2,0, j )
0 y2,1,j (1 2,1, j ) Y2,1, j y2,1,j (1 + 2,1,j )
0 y2,2, j (1 2,2, j ) Y2,2,j y2,2,j (1 + 2,2,j )
0 y2, , j (1 2, ,j) Y2, ,j y2, , j (1 + 2, , j )
0 y3,0,j (1 3,0, j ) Y3,0, j y3,0,j (1 + 3,0,j )

The following water volume balance equations are calculated and, accordingly, generate the pollutant mass conservation equations:
DX11 = X1,0 + X1,1 ± X1, X2,0 for single balance node 1
DX22 = X2,0 + X2,1 + X2,2 ± X2, X3,0 for single balance node 2
DX12 = X1,0 + X1,1 ± X1, + X2,1 + X2,2 ± X2, X3,0 for balance of nodes 1 to 2
Dq11, j = X1,0 Y1,0, j + X1,1 Y1,1, j ± X1, Y1, , j X2,0 Y2,0, j for single balance node 1
Dq22, j = X2,0 Y2,0, j + X2,1 Y2,1, j + X2,2 Y2,2, j ± X2, Y2, ,j X3,0 Y3,0, j for single balance node 2
Dq12, j = X1,0 Y1,0, j + X1,1 Y1,1, j + X2,1 Y2,1, j + X2,2 Y2,2, j ± X1, Y1, ,j ± X2, Y2, ,j X3,0 Y3,0, j for balance of nodes 1 to 2

The unknown values of the latent water quantities xi,λ of node i (where: i = 1 and 2) are given as follows:
x1, = ±(Dx11 x1,0 x1,1 + x2,0) for Dx11 = 0
x2, = ±(Dx22 x2,0 x2,1 x2,2 + x3,0 ) for Dx22 = 0

The latent term of the pollutant/tracer concentrations y1,λ,j is computed as follows (and the y2,λ,j accordingly):

Dq11, j x 1,0 y1,0, j x 1,1 y1,1, j + x 2,0 y2,0, j


y1, ,j =± x 1,
for Dq11, j = 0

Dq22, j x2,0 y2,0, j x 2,1 y2,1, j x2,2 y2,2, j + x3,0 y3,0, j


y2, ,j =± x2,
for Dq22, j = 0

Finally, all inequality constraints for cross-section (1,0), combining expressions (2) and (3) from the water volume balance, are given below;
accordingly, the corresponding inequality constraint sets are computed for the rest of the cross-sections:
0 x1,0 (1 1,0 ) DX11 (X1,1 ± X1, X2,0 ) x1,0 (1 + 1,0 )
0 x1,0 (1 1,0 ) DX12 (X1,1 ± X1, + X2,1 + X2,2 ± X2, X3,0 ) x1,0 (1 + 1,0 )

In addition, from the pollutant mass conservation equations for cross-section (1,0) and for each pollutant (j) the following inequality constraints
are computed:
0 x1,0 (1 1,0 ) y1,0,j (1 1,0,j ) Dq11, j (X1,1 Y1,1, j ± X1, Y1, ,j X2,0 Y2,0, j )
x1,0 (1 + 1,0 ) y1,0,j (1 + 1,0,j )
0 x1,0 (1 1,0 ) y1,0,j (1 1,0,j ) Dq12, j (X1,1 Y1,1, j + X2,1 Y2,1, j + X2,2 Y2,2, j ± X1, Y1, ,j ± X2, Y2, ,j X3,0 Y3,0, j )
x1,0 (1 + 1,0 ) y1,0,j (1 + 1,0,j )

Appendix B. Glossary including abbreviations and symbols

Dqab,j The residuals of the pollutant mass balance equations written for a combination of successive nodes (more precisely, from node i = a up to node i = b).
DVR Data validation and reconciliation
DXab The residuals of the water volume balance equations written for a combination of successive nodes (more precisely, from node i = a up to node i = b).
EC The electrical conductivity of water
F The objective function which is minimized and the corrected values of the parameters are derived from.
i The balance node with i = 1,2 …,K. Their enumeration starts from the river springs (i = 1, being the first node) towards river estuaries (i = K being the last node),
following the flow direction of the river.
(i,s) The measurement cross-sections, where the quantitative and qualitative measurements are undertaken. The enumeration of the cross-sections follows the flow direction
of the river. They are denoted by using two indices (i,s): the first one, i, is connected to the node enumeration (where: i = 1,2 …,K+1) and the second one, s, to the
enumeration of the cross-sections within node i (where: s = 0,1,2, …,ni and ni is the total number of tributaries at node i).
j The pollutants or tracers, which are taken into account into the methodology, with a j = 1, …,p
MCExi,s The measurement error values of discharges which were generated by Monte-Carlo simulations, assuming normal distribution.
MCEyi,s,j The measurement error values of concentrations, which were generated by Monte-Carlo simulations, assuming normal distribution.
MV Measured values
ND The normal distribution
NDH- Error Case 8: RE following the ND and high negative values of SE
NDH+ Error Case 7: RE following the ND and high positive values of SE
NDL- Error Case 4: RE following the ND and low negative values of SE

382
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

NDL+ Error Case 3: RE following the ND and low positive values of SE


NDM- Error Case 6: RE following the ND and medium negative values of SE
NDM+ Error Case 5: RE following the ND and medium positive values of SE
NDN Error Case 2: RE following the normal distribution (ND) and no SE
ni The total number of tributaries at node i
OV Optimized values (corrected by the proposed methodology)
p The total number of tracers with measured concentration.
qi,s,j The pollution load of a pollutant or tracer j at a cross-section (i,s) which is estimated as the product of the measured values of the water flow rate and the pollutant
concentration: xi,syi,s,j.
s It enumerates the cross-sections within node i with s = 0,1,2 …,ni. It starts from the cross-section at the main river where s = 0 and ends to the last tributary within the
node i where s = ni (where ni is the total number of tributaries at node i).
SE Systematic error
TDS The total dissolved solids concentration
UD The uniform distribution
UDN Error Case 1: RE following the uniform distribution (UD) and no SE
WLS Weighted least squares
xi,s The measured values of the river discharge at a cross-section (i,s) (i = 1,2 …,K+1; s = 0,1, …,ni,λ).
Xi,s The corrected/optimized value (derived from the proposed methodology) of the river discharge at a cross-section (i,s)
Xreali,s The real value of the river discharge at a cross-section (i,s)
yi,s,j The measured values of the concentration of a pollutant/tracer j at a cross-section (i,s) (i = 1,2 …,K+1; s = 0,1 …,ni,λ and j = 1, …,p).
Yi,s,j The corrected/optimized value (derived from the proposed methodology) of the pollutant concentration of a pollutant or tracer j at a cross-section (i,s)
Yreali,s,j The real value of the concentration of a pollutant or tracer j at a cross-section (i,s)
εi,s The absolute maximum relative measurement error of the river discharge at a cross-section (i,s)
ζi,s,j The absolute maximum relative measurement error of the concentration of a pollutant or tracer j at a cross-section (i,s)
λ In each single-balance node, we account for the existence of an unknown and not directly-measured water quantity (referred to as “latent”) in the process. The latent
discharge of a balance node is assumed to correspond to the runoff of a catchment area, which is included between all inflowing cross-sections and the outflowing cross-
section of the node. The latent term is signified by the Greek index s = λ

References Himmelblau, D.M., 1985. Material balance rectification via interval arithmetic. The use of
computers in chemical engineering. Process Systems Engineering 92, 121–133.
ISO 748, 1979. Liquid Flow Measurement in Open Channels –velocity Area Methods.
Allan, J.D., Reyeros de Castillo, M.M., Allen, J.D., 2007. Stream Ecology. Springer, International Organization for Standardization (ISO), Geneva, Switzerland.
Dordrecht, The Netherlands. ISO 1100, 1981. Liquid Flow Measurement in Open Channels: Part 1: Establishment and
APHA, 1999. Standard Methods for the Examination of Water and Wastewater, twentieth Operation of a Gauging Station and Part 2: Determination of Stage-Discharge
ed. American Water Works Αssociation, Water Environment Federation. Relation. International Organization for Standardization (ISO), Geneva, Switzerland.
Baloochy, B., Shokri, S., Marvast, M.A., 2010. A Fast Method for Data Validation in RTO ISO 772, 1988. Liquid Flow Measurement in Open Channels: Vocabulary and Symbols,
Technology. Int. J. Chem. Eng. Appl. 1 (3), 230. third ed. International Organization for Standardization (ISO), Geneva, Switzerland.
Bekri, E.S., Disse, M., Yannopoulos, P.C., 2012. Methodological framework for correction Kim, K., Lee, J.S., Oh, C.-W., 2002. Inorganic chemicals in an effluent-dominated stream
of quick river discharge measurements using quality characteristics. In: Yannopoulos, as indicators for chemical reactions and streamflows. J. Hydrol. 264 (1–4), 147–156.
P., Dimas, A. (Eds.), Paper Presented at 2nd Common Conference on Integrated Water Kinori, B.Z., Mevorach, J., 1984. Manual of surface drainage engineering. In: Stream Flow
Resources Management for Sustainable Development. Hellenic Hydrotechnical Engineering and Flood Protection. vol. II Elsevier Science, Amsterdam, The
Association and Greek Committee for Water Resources Management, Patras, Greece Netherlands.
(in Greek) 11-13 October 2012. Kirchner, J.W., Feng, X., Neal, C., 2000. Fractal stream chemistry and its implications for
Bekri, E.S., Yannopoulos, P.C., 2012. The interplay between the Alfeios River Basin contaminant transport in catchments. Nature 403, 524–527.
components and the exerted environmental stresses: a critical review. Water, Air, Soil Koelling, C., 2004. SIMK-Calibration of streamflow - gauging stations in rivers and canals.
Pollut. 223 (7), 3783–3806. In: Paper Presented at 5th International Conference on Innovation in Hydraulic
Bekri, E.S., Yannopoulos, P.C., Disse, M., 2013. A combined linear optimisation metho- Efficiency Measurement IGHEM. University of Applied Sciences, Lucerne,
dology for water resources allocation in an Alfeios river subBasin (Greece) under Switzerland.
uncertain and vague system conditions. In: Paper Presented at European Geosciences Kuehn, D.R., Davidson, H., 1961. Computer control. II. Mathematics of control. Chem.
Union General Assembly 2013, 22-27 April 2013, Vienna, Austria. Eng. Prog. 57, 44–47.
Carter, R.W., Anderson, I.E., 1963. Accuracy of current meter measurements. J. Hydraul. Kuriqi, A., Ardiçlioglu, M., 2018. Investigation of hydraulic regime at middle part of the
Div. 89 (4), 105–115. Loire River in context of floods and low flow events. Pollack Period. 13 (1), 145–156.
Delava, P., Maréchal, E., Vrielynck, B., Kalitventzeff, B., 1999. Modelling of a crude oil Kuusisto, E., 1996. Hydrological measurements. In: Water Quality Monitoring - A
distillation unit in term of data reconciliation with ASTM or TBP curves as direct Practical Guide to the Design and Implementation of Freshwater Quality Studies and
input – application crude oil preheating train. In: Paper Presented at ESCAPE-9 Monitoring Programmes. United Nations Environment Programme, E&FN Spon,
Conference, 31 May - 2 June 1999, Budapest, Hungary. London, UK.
Dickinson, W.T., 1967. Accuracy of discharge determinations. In: Hydrology Papers. Ford Langenstein, M., Jansky, J., Laipple, B., 2004. Finding Megawatts in nuclear power plants
Collins, Colorado, USA. with process data validation. In: Paper Presented at ICONE12, Nuclear Engineering
Eaton, A.D., Clesceri, L.S., Greenberg, A.E., 1995. Standard Methods for the Examination Division. vol. 2 25-29 April 2004, Arlington, Virginia, USA.
of Water and Wastewater, nineteenth ed. American Public Health Association, Le Coz, J., 2012. A Literature Review of Methods for Estimating the Uncertainty
Washington D.C., USA. Associated with Stage-Discharge Relations. WHO, Lyon, France.
Elhadi, N., Harrington, A., Hill, I., Lau, Y.L., Krishnappan, B.G., 1984. River mixing - a Lindo Systems Inc, 1996. LINDO User's Manual. Lindo Systems Inc, Chicago, Illinois.
state-of-the-art report. Can. J. Civ. Eng. 11 (3), 585–609. Mah, R., Stanley, G.M., Downing, D., 1976. Reconciliation and rectification of process
Fogel, E., Huang, Y.F., 1982. On the value of information in system identification- flow and inventory data. IEC Process Design Development 15 (1), 175–183.
bounded noise case. Automatica 18 (2), 229–238. Mandel, D., Abdollahzadeh, A., Maquin, D., Ragot, J., 1998. Data reconciliation by in-
Fujita, I., 2017. Discharge measurements of snowmelt flood by space-time image velo- equality balance equilibration: a LMI approach. International Journal of Mineral
cimetry during the night using far-infrared camera. Water 9, 269. https://doi.org/10. Processing 53 (3), 157–169.
3390/w9040269. Manariotis, I.D., Yannopoulos, P.C., 2004. Adverse effects on Alfeios river basin and an
Gartner, J.W., Ganju, N.K., 2007. Correcting acoustic Doppler current profiler discharge integrated management framework based on sustainability. Environ. Manag. 34 (2),
measurement bias from moving-bed conditions without global positioning during the 261–269.
2004 Glen Canyon Dam controlled flood on the Colorado River. Limnol Oceanogr. Maquin, D., Bloch, G., Ragot, J., 1991. Data reconciliation for measurements. Eur. J.
Methods 5 (6), 156–162. Diagnosis Saf. Autmation 1 (2), 145–181.
Hancock, J., 2013. Special Report - Next Generation Data Validation and Equation Based McMillan, H., Krueger, T., Freer, J., 2012. Benchmarking observational uncertainties for
Reconciliation Software Technology. Global Business Media, U.K. hydrology. Hydrol. Process. 26 (26), 4078–4111.
Herschy, R.W., 1971. In: The Magnitude of Errors at Flow Measurements Stations: Water Milanese, M., Belforte, G., 1982. Estimation theory and uncertainty intervals evaluation
Resources Board: Technical Report. Reading Bridge House, Reading, Berkshire, UK, in presence of unknown but bounded errors: linear families of models. IEEE Trans.
TN11. Autom. Control 27 (2), 408–413.
Herschy, R.W., 1985. Streamflow Measurement. E&FN Spon, Elsevier Applied Science, Milanese, M., Norton, J., Piet-Lahanier, H., Walter, E., 1996. Bounding Approaches to
London, UK. System Identification. Springer US, New York, USA.
Himmelblau, D.M., 1978. Fault detection and diagnosis in chemical and petrochemical Müller, A., 1988. Discharge and velocity measurements. In: Balkema, A.A. (Ed.),
processes. Chemical Engineering Monographs, vol. 8 Elsevier Scientific, New York, Proceedings of the Short Course on Discharge and Velocity Measurements,
USA.

383
E.S. Bekri, et al. Journal of Environmental Management 247 (2019) 371–384

International Association for Hydraulic Research, IAHR Proceedings no.2, 26-28 WFD, 2000. Water framework directive. Directive 2000/60/EC of the european parlia-
August 1987, Rotterdam, The Netherlands. ment and of the council of 23 october 2000 establishing a framework for community
Nakagawa, H., Ono, M., Oda, M., Nishijim, S., 2007. Development of a discharge mea- action in 12 the field of water policy. Off. J. Eur. Commun. L327, 1–72
surement technique combined with measurement of mean flow velocity and nu- 22.12.2000, 43.
merical simulation of cross-sectional velocity distribution; field verification in a large White, W.R., Mueller, A., 1988. Discharge measuring methods in open channels. Paper
river. J. Hydrosci. Hydraul. Eng. 25, 77–88. presented at Short course on discharge and velocity measurements. In: Balkema, A.A.
NCSU, 2008. Surface Water Flow Measurement for Water Quality: Monitoring Projects, (Ed.), IAHR Proceedings no.2, 26-28 August 1987, Rotterdam, The Netherlands.
TechNotes 3. National Nonpoint Source Monitoring Program, U.S.EPA., USA. International Association for Hydraulic Research.
Özyurt, D.B., Pike, R.W., 2004. Theory and practice of simultaneous data reconciliation WMO, 1980. Manual on stream gauging: vol. I - fieldwork and vol. II - computation of
and gross error detection for chemical processes. Comput. Chem. Eng. 28 (3), discharge. In: Operational Hydrology: Report No. 13. World Meteorological
381–402. Organization (WMO) - No. 519, Geneva, Switzerland.
Pelletier, P.M., 1988. Uncertainties in the determination of river discharge: a literature Yannopoulos, P.C., 1995. Bubble size and movement in wide channel flows. Paper pre-
review. Can. J. Civ. Eng. 15 (5), 834–850. sented at the 4th Greek national congress on mechanics, demokritus university of
Peters, N., 1994. Hydrologic processes. In: Biogeochemistry of Small Catchments: A Tool thrace. In: Theocaris, P.S., Gdoutos, E.E. (Eds.), Vol. II: Mechanics of Fluids and
for Environmental Research: Scientific Committee on Problems of the Environment Thermal Sciences, 26-29 June 1995, Xanthi, Greece (in Greek).
(SCOPE) of the International Council of Scientific Unions (ICSU) and of the United Yannopoulos, P.C., 2008. Development of low cost methodologies for quickly predicting
Nations Environment Programme (UNEP). John Wiley & Sons Ltd, New-York, USA and monitoring of river pollution: final Report. In: Pythagoras II - Environment
(Chapter 9). Project. OPEVTΙΙ – ESF, University of Patras, Patras, Greece (in Greek).
Podimata, M., Yannopoulos, P.C., 2013. Evaluating challenges and priorities of a trans- Yannopoulos, P.C., Bekri, E.S., 2010. Correction of quick discharge measurements in
regional river basin in Greece by using a hybrid SWOT scheme and a stakeholders' rivers using natural tracers. In: Christodoulou, Stamou (Eds.), Paper Presented at 6th
competency overview. Int. J. River Basin Manag. 11 (1), 93–110. International Symposium on Environmental Hydraulics ISEH. vol. 2. Taylor and
Ragot, J., Maquin, D., 2005. Data validation and diagnosis using interval analysis. In: Francis Group, London, UK, pp. 23–25 June 2010,.
Paper Presented at 3rd European Workshop on Advanced Control and Diagnosis, 17- Yannopoulos, P.C., Demetracopoulos, A.C., Hadjitheodorou, C., 2008. Quick method for
18 November 2005, Mulhouse, France. open channel discharge measurements using air bubbles. J. Hydraul. Eng. 134 (6),
Ragot, J., Maquin, D., Bloch, G., 1991, July. Data reconciliation for process flow. In: 843–846.
European Control Conference, ECC'91. vol. 2. pp. 1430–1435. Yannopoulos, P.C., Manariotis, I.D., Ziogas, A.I., Kaleris, V.K., 2007. Methodology of river
Rantz, S.E., 1982. Measurement and computation of streamflow: volume 1: measurement pollution assessment and preliminary results. In: Di Silvio, G., Lanzoni, S. (Eds.),
of stage and discharge. In: USGE Water-Supply Paper 2175, 01-01. U.S. Government Paper Presented at the 32nd IAHR Congress Harmonizing the Demands of Art and
Printing Office, Washington D.C., USA. Nature in Hydraulics, 1-6 July 2007, Venice, Italy.
Sauer, V.B., Meyer, R.W., 1992. Determination of error in individual discharge mea- Yannopoulos, P.C., Mavrikos, G., Demetracopoulos, A., Hatzitheodorou, C., 2000.
surements. In: Open File Report 92-144. U.S. Geological Survey, Norcross, Georgia, Velocity measurement in channels using a sphere. In: Christodoulou, G., Stamou, A.,
USA. Nanou, A. (Eds.), Paper Presented at 8th National Congress of the Hellenic
Schmidt, C., Musolff, A., Trauth, N., Vieweg, M., Fleckenstein, J.H., 2012. Transient Hydrotechnical Union. Hellenic Hydrotechnical Union, Athens, Greece (in Greek) 19-
analysis of fluctuations of electrical conductivity as tracer in the stream bed. Hydrol. 21 April 2000.
Earth Syst. Sci. 16 (10), 3689–3697. Yorke, T.H., Oberg, K.A., 2002. Measuring river velocity and discharge with acoustic
Schrage, L.E., 1997. Optimization Modeling with LINDO, fifth ed. Duxbury Press, Pacific Doppler profilers. Flow Meas. Instrum. 13 (5–6), 191–195.
Grove, USA. Ziabras, T., Tasias, S., 1992. River Discharge Measurements through Natural Tracers in
Simpson, M.R., 2001. Discharge measurements using a broad-band acoustic Doppler Alfeios River. Diploma thesis. Environmental Engineering Laboratory, Civil
current profiler. In: U.S. Geological Survey Open-File Report 01-01. U.S. Dept. of the Engineering Department, University of Patras, Greece (in Greek).
Interior, U.S. Geological Survey, Sacramento Calif., Denver CO, USA.

384

You might also like